The abundance trap

A reader can stop paying. That sounds obvious, but it changes the entire problem.

A school can compel attendance. A company can require a report. A newspaper may continue publishing because advertising, subscriptions, institutional contracts or ownership structures support it. A platform like Substack depends much more directly on a fragile act of confidence.

Someone decides, I want more from this person. Then, sometimes, they make another decision: I am willing to pay for it.

That transaction contains more than access to words. Words are already abundant. Information is abundant. Opinion is abundant. Artificial intelligence makes all three more abundant still. The payment is closer to a judgment:

This particular source is worth returning to.

That is why the problem of AI-generated abundance is not simply a question of quality. It is a question of trust, attention and economic value.

When everything looks publishable

Imagine opening a platform where ten thousand new essays appeared this morning. Many are competent. Some are excellent. Most have attractive titles. Nearly all are grammatically clean. They contain quotations, historical comparisons, personal openings, objections and thoughtful conclusions.

But they did not all come into existence in the same way.

One may have grown from twenty years of professional experience. Another from months of reading. Another from a genuine conversation between a writer and an AI system that challenged the original argument. And another may have begun twenty minutes earlier with: Give me something interesting to publish today.

From the reader’s side, all of them arrive through the same door.

That is what abundance changes. When supply is limited, discovery is largely about finding enough good material. When supply becomes effectively infinite, discovery becomes about avoiding the cost of examining everything else.

The problem is no longer publication. It is selection.

We have seen this economics before

YouTube learned this early. Once anyone could upload video, the problem stopped being whether enough material existed. More video was being uploaded than any person could watch in many lifetimes, so the valuable system increasingly became the one deciding what appeared next.

TikTok pushed the logic further. The supply of short video became so abundant that the individual object barely needed a future. Swipe, another video. Swipe again, another. The platform did not need you to remember what you watched yesterday; it needed the next object to be interesting enough to prevent you from leaving now.

That is a particular economics of abundance:

the individual object becomes disposable while the feed becomes permanent.

A writing platform should be careful before adopting the same logic, because writing depends unusually heavily on a relationship between the individual piece and what came before it. A writer returns to questions. A reader remembers an earlier essay. An argument continues. A contradiction becomes visible because six months ago the same person said something different.

The archive matters because the person matters, and time matters because the relationship accumulates through it.

That is precisely what becomes weaker when writing turns into another high-volume feed.

The pressure to remain visible

An essay begins to compete not merely with other essays, but with an expectation of perpetual novelty: something new this morning, another thought this afternoon, a Note about the thought, a response to somebody else’s Note, a reaction to the news, a follow-up tomorrow.

The platform may never explicitly demand any of this. The economics can.

Visibility rewards activity. Activity produces visibility. Visibility produces subscribers. Subscribers create pressure to remain visible. That cycle existed before AI, but AI removes one of its last constraints.

The exhausted writer no longer has to choose between publishing something weak and publishing nothing. The machine offers a third option: publish something competent.

Once competence becomes available on demand, silence begins to look economically irrational.

That is where the problem becomes more interesting. Not because every frequent writer is shallow, and certainly not because every AI-assisted article is empty. The danger is structural: a system can begin rewarding the person who produces the appearance of intellectual continuity more consistently than the person who actually waits for thought to mature.

The slow writer may spend three weeks discovering that the original premise was wrong and publish one article. The fast publisher may ask a model for twenty variations on the premise and publish twelve. If both are competing for the same finite attention, production velocity begins to matter even when everyone involved claims to value depth.

Readers adapt too

Readers do not remain unchanged inside this system. They skim more, trust headlines less, open fewer emails and become more selective about subscriptions. One paid newsletter feels harmless. Ten begin to feel expensive. Twenty become impossible to read.

And beneath all of that sits a new question:

Why am I paying for another explanation of something I can ask an AI to explain to me now?

That question does not eliminate the paid writer. But it changes what the writer must be worth paying for.

Generic explanation becomes harder to monetize when explanation itself is abundant. Generic commentary becomes harder to monetize when commentary can be generated instantly. Generic elegance becomes less valuable when elegant prose is cheap. Even a competent synthesis of publicly available information begins to lose scarcity.

The reader does not need another article.

The reader needs a reason to care about this one.

The economic unit may be moving from the article to the person

What creates that reason?

It may be expertise, judgment, experience, style, access or trust. It may be a particular way of noticing something everyone else missed. Usually it will be some combination accumulated over time.

But whatever the combination is, something important has changed.

The economic unit quietly moves away from the individual article and toward the person behind it.

That sounds paradoxical in an age when AI makes producing articles easier, but perhaps it follows directly from abundance. When words are scarce, the words themselves have value. When competent words become abundant, the source matters more.

A subscription increasingly says:

I trust this person’s selection, judgment and intellectual continuity enough to return.

That is very different from paying simply because someone can produce readable prose.

The abundance trap

This is where Substack’s business model becomes entangled with the cultural problem.

If readers increasingly encounter polished but interchangeable writing, the perceived value of subscribing falls. If perceived value falls, fewer readers pay. If fewer readers pay, writers feel more pressure to grow. Growth requires visibility, visibility rewards activity, and activity becomes easier with AI.

The response to declining value can therefore become more of the thing contributing to the decline: more posts, more Notes, more analysis, more reactions, more apparent insight and more intellectual presence.

The platform receives more content and risks becoming less valuable. The writer produces more and risks becoming less distinct. The reader receives more and becomes less willing to pay.

That is the abundance trap.

Nothing has to collapse dramatically. Quality can erode through dilution rather than destruction. The good writing remains; it is simply surrounded by enough plausible writing that finding it becomes work.

When discovery becomes the product

Once that happens, Substack faces a temptation familiar to every large platform: solve discovery algorithmically.

Rank, recommend, predict. Show the reader what is most likely to keep the reader engaged.

That can work. It can also move the platform closer to the feed it once seemed to offer an escape from.

The paradox is sharp. A platform built around direct relationships between writers and readers may need increasingly powerful mediation simply because too many writers can now produce too much material.

The gatekeeper returns, not at publication but at attention.

And perhaps some form of gatekeeping is inevitable. No system containing more material than a person can inspect can avoid selection. The important question is therefore not whether selection happens, but what the selection optimizes for.

Frequency, engagement, growth, payment, time spent, reader retention, reputation or long-term satisfaction all produce different outcomes. Those choices are not neutral because they determine what kind of writer the platform rewards.

If visibility follows volume, AI-assisted volume has an advantage. If visibility follows outrage, a different pathology wins. If visibility follows existing popularity, new writers struggle to enter. And if the platform attempts to judge quality directly, it risks becoming the editorial authority it once helped writers circumvent.

There is no clean solution.

The obvious solution does not work either

Substack cannot simply say: Fine. We will protect quality by restricting AI.

The writer using AI to translate an original argument into English may be doing exactly the kind of work the platform should want more of. The journalist using it to search a thousand pages of public records may be doing better reporting. The scientist using it to make technical knowledge readable may be widening access. A disabled writer may use it to overcome a barrier that previously made writing difficult or impossible.

A thoughtful writer may also use AI as an adversary, asking it to attack an argument until the weaknesses become visible. In all of these cases, the machine can strengthen the relationship between the writer and the work rather than weaken it.

A blunt restriction would therefore protect the platform from some low-value production by excluding some of the people AI most usefully empowers.

The distinction cannot simply be AI or no AI.

The harder distinction is between AI that helps a person express, investigate or challenge something that genuinely exists upstream, and AI used to manufacture the appearance that something existed upstream when it did not.

From the finished page alone, those two processes may be extremely difficult to distinguish.

Two kinds of openness

Substack is therefore caught between two forms of openness.

The old one was simple:

Anyone should be able to publish.

That removed permission.

The new one is different:

Anyone can now produce publishable material at enormous scale.

That removes friction.

They sound similar, but they are not.

The first solved an old problem of gatekeeping. A writer no longer needed an editor, publisher or institution to decide whether the public was allowed to encounter the work. That was a remarkable change.

AI introduces another one. The same writer can now produce far more material, far more quickly, at a level of basic competence that once required much more time.

So can everyone else.

The old gatekeeper asked: Who is allowed to speak? The open platform answered: Anyone.

AI adds: And they can speak almost without stopping.

At that point, the difficult question is no longer access.

It is attention.

What exactly is worth paying for?

Who is worth reading, returning to and paying?

These may become the defining questions of writing platforms in the AI era, and they cannot be answered simply by asking whether a machine touched the prose.

A writer may use AI heavily and still have something worth returning to. Another may type every word personally and have very little to say.

The scarce thing is not necessarily human keystrokes.

It may be intellectual distinctiveness: a recognizable judgment, a body of experience, a history of being right, wrong, uncertain and willing to correct course, or a relationship between what the writer publishes today and what the same writer believed six months ago.

It may simply be a mind that the reader gradually learns how to trust.

That takes time.

And perhaps time becomes more valuable precisely because AI reduces the time required to manufacture the appearance of it.

The reader becomes the final gatekeeper

This brings us back to the simplest fact.

A platform can recommend, a writer can publish, an algorithm can rank and a model can generate. But none of them can force the final transaction.

The reader still decides:

I want more from this person.

And then, occasionally:

I am willing to pay for it.

If Substack tries too aggressively to decide who deserves that attention, it risks becoming the gatekeeper it originally helped writers escape. If it refuses to make any distinctions at all, the burden moves almost entirely to the reader.

And the reader possesses one extremely powerful tool that requires no AI detector, no watermark, no policy and no algorithm:

they can leave.

Perhaps that is the discipline abundance eventually imposes. When words become almost unlimited, the value of producing more words falls, while the reason to return becomes more scarce.

And that may be the real economic question underneath the entire debate:

In a world where almost anyone can produce something worth reading once, who can remain worth returning to?

Karim Spring